These weeknotes seem to be settling into a nice bi-weekly rhythm – I started to write them last week but realised I didn’t have a lot to write about, and overcompensated with too much detail on the few things I did. Holding off for a week seems like a good move – I’ve got a more varied set of stuff to write about now which seems more conducive to useful reflection. So, without further ado:
I was in London for a few days last week, mostly to say goodbye to my good friend Al who is moving to the US, but it also gave me a chance to catch up with some folks I’m working with in person. Marco from Retechnica and I hatched some plans for a small prototyping / exploration project we’re going to start next week (and caught up in real life for the first time in years), and I finally met the whole crew from Adia in person for the first time when I visited them to deliver a workshop on developing product ideas that use machine learning. It was the first time i’d delivered this workshop (Although it’s based on a few techniques I used a lot at BBC R&D), and it got a really positive reception.
The workshop had three related aims:
- To give enough of an introduction to the practicalities of ML for non-specialists (product and design people, in particular, as well as non-specialist developers) to start thinking about how (and indeed, if) it could be used to add value in their projects
- To generate some ideas of ML-based features that might fit within their own projcts and strategy (kept thoroughly grounded in user research, and real user needs, of course)
- To use these ideas to start thinking through some of the trickier questions around the use of ML and its possible consequences – measuring and understaning error, the potential for bias, the need for fairness and transparency, and more.
The overall aim of the workshop is to get product teams thinking about ML in practical concrete terms as a tool that can be used to solve real problems (rather than as part of an abstract ‘data and AI strategy’), and to give some early awareness of some of the trickier challenges associated with using it in a way that’s fair, just, and useful. I’m really pleased with how it went and would like to try out the format with more people – please get in touch if this is something that might be useful to you (in particular: if you’re a charity or non-profit with values approximately aligned to my own, i’ll happily run this for free if you’re in Barcelona, or for (rail) travel costs only elsewhere in Europe).
This week has been spent working mostly with Adia again, getting stuck into another round of product development. I’ve also spent time working with Jakub to prepare our presentation for Eufonic Urba, coming up next week.
Yesterday I attended the Ctrl-Z AI Zine Fair, a ‘fringe’ event of the ACM conference on Fairness Accountability and Transparency which was held here this year. Ctrl-Z was a wonderful experience – thought-provoking workshops on dreaming and speculative futures, Cybernetic racism, and Classification as divination. I also picked up a haul of really thought provoking critical writing on technology in ‘zine form – the absolute standout being Sophie Wang’s ‘Science Under the Scope‘ – a wonderful introduction to Strong Objectivity and the social study of science, in comic form. It was also a real treat to see Andrew Nicolau who’d come to visit for the zine fair, and with whom I spent a lovely evening catching up.
Aside from that i’ve been messing around with the YOLO9000 object detection model and the instagram API for a nacent project on ML bias, community and geography which i’ll write more about next time:
- “How Art Holds AI to Account” by Daphne Miller
- Oz Keyes amazing piece for Logic magazine on gender, computer vision, and how, in general surveillance and measurement systems shape and discipline the people they survey: “The Body Instrumental”
- David Beer’s piece on “Twitter and Facial Recognition” from his weekly newsletter which I highly recommend subscribing to.